rereduplication

An interesting feature of the Indo-European verbal (and sometimes nominal) system is so called 'reduplication'.

Reduplication in Indo-European is the repetition of the first consonant plus an extra vowel before the action root.

A famous example is the Indo-European word for 'wheel' one of the few nominal roots with reduplication:
*kʷe-kʷlo-
Note that the reduplication is solely the first consonant, never the cluster. Well never is, of course, a big word. It does seem to happen sometimes.

Typically in Indo-European we reconstruct two types of reduplication, one with a vowel *e after the reduplicated consonant, one with the vowel *i. Obviously we like to reconstruct some kind of earlier **ə that gave both reflexes, but in the latest phase of PIE these two were definitely there.

Are these reconstructions really that natural though? Maybe not. Many languages seem to indicate a different system. For once Sanskrit was not authoritative in the opinions of the earlier Indo-Europeanists (typically, when there's doubt, the Sanskrit form is considered correct). No, this time it's Greek. Greek knows solely two types of reduplication. i-reduplication in presents (and very rare aorists), and e-reduplication in the perfect.

When we have a look at three other languages with attest reduplication, we get a very different view though. The languages I will discuss are Gothic, Sanskrit and Tocharian B. I've been told that Latin also displays some non-standard reduplication, but since my Latin isn't much better than 'dulce et decorum est pro patria mori', I'll leave that languages for people with more expertise.

First, let's talk about Gothic. Gothic is a nice language, for it displays reduplication often, and consistently throughout it's full history (which is awefully short, but still), unlike any other Germanic language. There's some proof of reduplication in other languages (I believe an odd Old-English form for 'rowed' was 'roard' (though spelling is probably wrong)). The other for is what we see in the preterite of do  did < *deda. But the *deda discussion is a different one all together where no consensus has been found yet.

Gothic reduplication, occurs in the preterite. This is exactly the way we like it in Indo-European reconstructions. There is one main issue though. The vowel use for reduplication is <ai> which phonetically is [ɛ] (some people are pro diphthong, especially in the long variant, I disagree, it has to be a monophthong).

This is not the vowel we expect in reduplication. PIE *e consistently became Gothic i in all contexts. Later though i was changed to ai in several contexts (this all happened before our first Gothic textual attestation).
 *i > ai/_r,h,ƕ This is linguist speak for 'e became <ai>  before r,h and ƕ (Similarly als *u > au/_r,h,ƕ for you symmetry lovers).

Not all reduplication verbs start with r,h or ƕ though. In fact I can't think of any. So a verb like letan you'd expect to have a preterite **lilot. What we actually have is lailot. Some people have desperately tried to explain this to say that the vocalism is analogy from the reduplicating verbs that do start with r,h and ƕ. Personally I do not find this very convincing. Nevertheless, it's very hard to have any <ai>  in Gothic outside of the above mentioned  position. Ony if it were the long variant of <ai> (not visible in writing), we would be able to have that vowel outside of that position.

If <ai> in reduplicating verbs is not from earlier i, we would expect it to be from an earlier *ai < PIE *oi, which is not at all convincing either. Because of this (and several other reasons), I think there might be reason to assume that reduplication in Gothic is not a direct descendant of Indo-European reduplication (this assumption is going to academically kill me :P),

If it is old though, we have either a very odd Indo-European reduplication, or a very odd development in Gothic. You choose ;).

Now, let us discuss Sanskrit. Sanskrit uses reduplication in much the same contexts as Greek. It's used in some presents/aorists (which, I believe, use exclusively the i-vocalism). Besides that the i-vocalic reduplication is used in desideratives and intensives (some indication of this, is also, and solely available in Greek (even more proof for my pro-Graeco-Iranian theory :P)). In the perfect though, we see a different situation all together. We don't see the velar-palatalising a that we would expect as a reflex of PIE *e. No, it reduplicates the sonorant semi-vowel or vowel of the root in the zero grade.

What does that mean? Here let me show you:
Stem: puṣ- present: puṣyati perfect: pupoṣa (< Vedic pupauṣa). So it actually reduplications the u. This is odd in many ways, especially because the sound u is not actually immediately following the first consonant when the verb is in the perfect. In the perfect plural it is though, so likely it got it from there.
This reduplication is also seen when the semi-vocalic consonant is an i. Also the a is reduplicated when it's in the zero grade of the root, or if the second consonant is a vocalic r. But to be honest, this might be the Indo-European *e, especially since it palatalises the reduplicated velars.

Note that this semi-vocalic root vowel even crosses cluster boundaries. So the root śru- gives śuśro-.

Another distinctive feature of Sanskrit reduplication is the way the consonants are treated. When a voiceless non-aspirate is reduplicated, the same consonant remains. An example was given above.

If a voiced aspirate and h < *gh, is aspirated, a voiced non-aspirate. For example
dhāv- > dadhāv
This is not that different from Greek, it's Grassmann's law. Only Greek has voiceless aspirates, and they thus give the voiceless non-aspirate. (τιθημι).

But in some roots, and this is really weird, a voiceless consonant reduplicated sometimes gives the most bizarre reflexes. One that comes to mind right now is the verb pibati 'he drinks' with the root pā-. That's right, the root shortened it's ā to a, then the root consonant became b while the reduplicative consonant remains p. Absolutely insane, but very real nevertheless. This is but a minority though. Most voiced reduplicated consonant remain voiced.

This is, once again, a very different form of reduplication than the one we see in Greek. I've been told that something similar happens in the few Latin verbs that have reduplication. But someone else is going to have to give examples.

Now we get to the last, and really crazy part. That's right, Tocharian B. Tocharian B (and probably also Tocharian A :P) is always crazy. It fails to conform to any type of normal behaviour. I love it.

Tocharian B has reduplication in only one class. The Preterite Participle.

The most normal Preterite Participle is perfectly normal. CäCaw (a<accented ä). Example:lätaw from the root lät-
The second group is C1eC1aw (a from accented ä). Example śeśaw. Though this looks like a perfectly normal reduplication, this is already bizarre. Because TochB e does not come from IE *e, rather it comes from IE *o. This o is also nowhere to be found in the root. śä-
The third group is C1eC1aC2äw (a from accented ä), For example lyelyakäw from root läk-
Next group is C1eC1éC2äw Example:  teteku (u = äw) from täk-, At last here we find a decent reduplication of the root vowel.
(j)VCäw where V is accented. For example: yáku root: yäk-
C1aC1áC2äw for example: lalālu (ā = accented a) from root: la-. TochB. a comes from Indo-European *h2e.
The next class is rather normal (no reduplication!) CäCáu for example: krämpau root: krämp-
The before last class is C1aC1áC2au for example kakārpau root: korp-
The last class I don't have any example of, but luckily isn't reduplicating either. áCau.

So in Tocharian we have reduplication with o, or h2e. And not once with *e or *i. Nor is the root semi-vowel reduplicated as we'd hope to see considering Sanskrit.

So there we have it! Three completely different reduplication systems, which are opposed directly against the Greek and reconstructed Indo-European form. Why we are still reconstructing the reduplication the way we are is actually a bit mysterious. It's the easiest, that's for sure. All these systems might have also undergone some analogy, especially Sanskrit and Gothic.

This leaves us with Tocharian B though, which went about being it's own crazy self (he likes to do that). Now we might want to mention that Tocharian split of from the Indo-European family well before the rest of the languages started branching off (but well after Anatolian branched off). Maybe this was indeed the crazy reduplication system PIE used to have, but then all reduplication was reduced to a schwa.

One thing is sure, the last word has definitely not been spoken about Indo-European reduplication, it's no where near as straight forward as some people like to think (Though in all honesty, to keep your sanity, I can imagine you'd rather think it's that simple than look at Tocharian; What can I say? I'm a linguistic masochist :P).